Journals should cap each author’s submissions – but with a twist

Having recently completed my editor-in-chief’s role for a management journal, I wrote my final editorial in appreciation of the care I witnessed when authors, reviewers and editors interacted in the production of knowledge.

It was often a treat to see how people conversed with people, how people listened to people, to resolve doubts and disagreements through clarifications. These interactions reminded me that knowledge production is a social accomplishment – as researchers, we do not only exchange our perspective and findings but also think about what these mean for social and organisational practices. However, this process risks being drastically eroded by the widespread offloading of writing, reviewing and even editorial decision-making to artificial intelligence.

AI advocates routinely argue that the ever-greater use of AI in the research process is inevitable. Yet failing to interrogate this narrative risks augmenting the existing crisis of credibility in some fields, including management.

Judged against empirical evidence, that crisis is real. In 2025 alone, more than 140,000 references to non-existent papers were calculated to have infiltrated scientific records. And while AI helps researchers to produce more papers more quickly (with direct effects on individual careers), the quality of scholarship is decreasing, as indicated through a measure of language complexity.

Further, many journals argue for human “guard rails” to check the accuracy of AI outputs, but empirical studies suggest that AI biases (caused by human biases in the training data) transfer to more human users, and then these biases get amplified through epistemic feedback loops in human-AI interactions. AI’s tendency to reproduce its training data also exacerbates the pre-AI decline in disruptive and innovative research.

Finally, many authors neglect to disclose AI use, even when journals’ well-intentioned policies insist that authors should do so – or that they shouldn’t use AI at all. Hence, such policies may be working less well than is assumed. AI use, in the end, is probably unpoliceable.

In response, I propose two policy moves for journals – one qualitative, one quantitative.

Qualitatively speaking, I propose a set of criteria for preparing submissions (as authors) and evaluating them (as editors and reviewers) that promote the distinct, unique voice of authors in terms of their mastery of the literature, method, analysis, contribution and recommendations for practice. This is a matter of paramount importance when the defining scarcity of our time is not access to information or even knowledge, but rather the capacity to develop and apply sound judgement in the presence of seemingly boundless AI slop.

Judging under uncertainty and justifying oneself before others is effortful, and developing that judgement is – or should be – a career-long process. The struggle matters because sound judgement matters. Possessing it is a central part of what it means to be an academic at all.

Perhaps at this point in technological history an academic can benefit from being able to use AI for some research tasks; a Socratic sparring partner could help develop ideas, for instance. But the ability to use the technology judiciously depends on decades of building expertise before AI’s arrival. Junior colleagues and PhD students may not find the conditions in which to develop such expertise, especially under current incentive systems that value number of publications over quality and relevance of ideas. Moreover, as expertise levels will likely suffer in future, the conditions for rebuilding expertise will shrink. This should concern us all.

Yet, none of our efforts to prevent this decline will be workable unless we also manage the current deluge of journal submissions, giving reviewers and editors the breathing space to write quality letters and reviews based on the proposed new set of evaluative criteria. Therefore, my quantitative proposal is to introduce submission caps.

Subject to amendments to suit the needs of a particular journal, I propose that each author be permitted to submit up to three new submissions per year to a journal (using ORCID numbers as an ID check). But there is a twist. To discourage the use of the journal as a preliminary check on a manuscript’s submission-readiness, I would add an incentive to improve the quality of first submission. Namely, for each invitation to revise and resubmit a new submission, each author would be allowed to submit one additional paper per year.

I believe that my proposals would also have the benefits of promoting equitable access to editorial resources, reducing congestion in the peer review pipeline and stemming the mounting volume of low-quality content permeating academic publishing, which will only intensify in the age of AI. In addition, it would help secure the conditions that promote trust in published works in the age of AI, with epistemic agency flowing back from Big Tech to journals and editors.

Some colleagues may baulk at a perceived limit to their productivity. However, let’s appreciate that even if a prolific scholar only managed to publish three articles in each of the best five journals in their field per year, that tally of 15 articles would hardly be detrimental to their career prospects.

Personally, I opt for AI-free scholarship owing to my interest in the social conditions that we create through our research. But I know others will go on using it. Some have even developed AI research tools that are commercially available. The key point is that if we can develop and assess human judgement in the research process, the impossibility of policing AI use will matter much less.

Dirk Lindebaum is professor of management and organisation at the University of Bath and owner of “AI-Free Scholarship”, a registered (non-commercial) trademark that applies to all his writing and teaching.